This repository has been archived by the owner on Apr 22, 2024. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 55
/
vaGTAO.hlsl
159 lines (139 loc) · 9.84 KB
/
vaGTAO.hlsl
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Copyright (C) 2016-2021, Intel Corporation
//
// SPDX-License-Identifier: MIT
///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
//
// XeGTAO is based on GTAO/GTSO "Jimenez et al. / Practical Real-Time Strategies for Accurate Indirect Occlusion",
// https://www.activision.com/cdn/research/Practical_Real_Time_Strategies_for_Accurate_Indirect_Occlusion_NEW%20VERSION_COLOR.pdf
//
// Implementation: Filip Strugar ([email protected]), Steve Mccalla <[email protected]> (\_/)
// Version: (see XeGTAO.h) (='.'=)
// Details: https://github.com/GameTechDev/XeGTAO (")_(")
//
// Version history: see XeGTAO.h
///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
#include "vaShared.hlsl"
#include "vaNoise.hlsl"
#ifndef __INTELLISENSE__ // avoids some pesky intellisense errors
#include "XeGTAO.h"
#endif
cbuffer GTAOConstantBuffer : register( b0 )
{
GTAOConstants g_GTAOConsts;
}
#include "XeGTAO.hlsli"
// input output textures for the first pass (XeGTAO_PrefilterDepths16x16)
Texture2D<float> g_srcRawDepth : register( t0 ); // source depth buffer data (in NDC space in DirectX)
RWTexture2D<lpfloat> g_outWorkingDepthMIP0 : register( u0 ); // output viewspace depth MIP (these are views into g_srcWorkingDepth MIP levels)
RWTexture2D<lpfloat> g_outWorkingDepthMIP1 : register( u1 ); // output viewspace depth MIP (these are views into g_srcWorkingDepth MIP levels)
RWTexture2D<lpfloat> g_outWorkingDepthMIP2 : register( u2 ); // output viewspace depth MIP (these are views into g_srcWorkingDepth MIP levels)
RWTexture2D<lpfloat> g_outWorkingDepthMIP3 : register( u3 ); // output viewspace depth MIP (these are views into g_srcWorkingDepth MIP levels)
RWTexture2D<lpfloat> g_outWorkingDepthMIP4 : register( u4 ); // output viewspace depth MIP (these are views into g_srcWorkingDepth MIP levels)
// input output textures for the second pass (XeGTAO_MainPass)
Texture2D<lpfloat> g_srcWorkingDepth : register( t0 ); // viewspace depth with MIPs, output by XeGTAO_PrefilterDepths16x16 and consumed by XeGTAO_MainPass
Texture2D<uint> g_srcNormalmap : register( t1 ); // source normal map (if used)
Texture2D<uint> g_srcHilbertLUT : register( t5 ); // hilbert lookup table (if any)
RWTexture2D<uint> g_outWorkingAOTerm : register( u0 ); // output AO term (includes bent normals if enabled - packed as R11G11B10 scaled by AO)
RWTexture2D<unorm float> g_outWorkingEdges : register( u1 ); // output depth-based edges used by the denoiser
RWTexture2D<uint> g_outNormalmap : register( u0 ); // output viewspace normals if generating from depth
// input output textures for the third pass (XeGTAO_Denoise)
Texture2D<uint> g_srcWorkingAOTerm : register( t0 ); // coming from previous pass
Texture2D<lpfloat> g_srcWorkingEdges : register( t1 ); // coming from previous pass
RWTexture2D<uint> g_outFinalAOTerm : register( u0 ); // final AO term - just 'visibility' or 'visibility + bent normals'
// Engine-specific normal map loader
lpfloat3 LoadNormal( int2 pos )
{
#if 1
// special decoding for external normals stored in 11_11_10 unorm - modify appropriately to support your own encoding
uint packedInput = g_srcNormalmap.Load( int3(pos, 0) ).x;
float3 unpackedOutput = XeGTAO_R11G11B10_UNORM_to_FLOAT3( packedInput );
float3 normal = normalize(unpackedOutput * 2.0.xxx - 1.0.xxx);
#else
// example of a different encoding
float3 encodedNormal = g_srcNormalmap.Load(int3(pos, 0)).xyz;
float3 normal = normalize(encodedNormal * 2.0.xxx - 1.0.xxx);
#endif
#if 0 // compute worldspace to viewspace here if your engine stores normals in worldspace; if generating normals from depth here, they're already in viewspace
normal = mul( (float3x3)g_globals.View, normal );
#endif
return (lpfloat3)normal;
}
// Engine-specific screen & temporal noise loader
lpfloat2 SpatioTemporalNoise( uint2 pixCoord, uint temporalIndex ) // without TAA, temporalIndex is always 0
{
float2 noise;
#if 1 // Hilbert curve driving R2 (see https://www.shadertoy.com/view/3tB3z3)
#ifdef XE_GTAO_HILBERT_LUT_AVAILABLE // load from lookup texture...
uint index = g_srcHilbertLUT.Load( uint3( pixCoord % 64, 0 ) ).x;
#else // ...or generate in-place?
uint index = HilbertIndex( pixCoord.x, pixCoord.y );
#endif
index += 288*(temporalIndex%64); // why 288? tried out a few and that's the best so far (with XE_HILBERT_LEVEL 6U) - but there's probably better :)
// R2 sequence - see http://extremelearning.com.au/unreasonable-effectiveness-of-quasirandom-sequences/
return lpfloat2( frac( 0.5 + index * float2(0.75487766624669276005, 0.5698402909980532659114) ) );
#else // Pseudo-random (fastest but looks bad - not a good choice)
uint baseHash = Hash32( pixCoord.x + (pixCoord.y << 15) );
baseHash = Hash32Combine( baseHash, temporalIndex );
return lpfloat2( Hash32ToFloat( baseHash ), Hash32ToFloat( Hash32( baseHash ) ) );
#endif
}
// Engine-specific entry point for the first pass
[numthreads(8, 8, 1)] // <- hard coded to 8x8; each thread computes 2x2 blocks so processing 16x16 block: Dispatch needs to be called with (width + 16-1) / 16, (height + 16-1) / 16
void CSPrefilterDepths16x16( uint2 dispatchThreadID : SV_DispatchThreadID, uint2 groupThreadID : SV_GroupThreadID )
{
XeGTAO_PrefilterDepths16x16( dispatchThreadID, groupThreadID, g_GTAOConsts, g_srcRawDepth, g_samplerPointClamp, g_outWorkingDepthMIP0, g_outWorkingDepthMIP1, g_outWorkingDepthMIP2, g_outWorkingDepthMIP3, g_outWorkingDepthMIP4 );
}
// Engine-specific entry point for the second pass
[numthreads(XE_GTAO_NUMTHREADS_X, XE_GTAO_NUMTHREADS_Y, 1)]
void CSGTAOLow( const uint2 pixCoord : SV_DispatchThreadID )
{
// g_samplerPointClamp is a sampler with D3D12_FILTER_MIN_MAG_MIP_POINT filter and D3D12_TEXTURE_ADDRESS_MODE_CLAMP addressing mode
XeGTAO_MainPass( pixCoord, 1, 2, SpatioTemporalNoise(pixCoord, g_GTAOConsts.NoiseIndex), LoadNormal(pixCoord), g_GTAOConsts, g_srcWorkingDepth, g_samplerPointClamp, g_outWorkingAOTerm, g_outWorkingEdges );
}
// Engine-specific entry point for the second pass
[numthreads(XE_GTAO_NUMTHREADS_X, XE_GTAO_NUMTHREADS_Y, 1)]
void CSGTAOMedium( const uint2 pixCoord : SV_DispatchThreadID )
{
// g_samplerPointClamp is a sampler with D3D12_FILTER_MIN_MAG_MIP_POINT filter and D3D12_TEXTURE_ADDRESS_MODE_CLAMP addressing mode
XeGTAO_MainPass( pixCoord, 2, 2, SpatioTemporalNoise(pixCoord, g_GTAOConsts.NoiseIndex), LoadNormal(pixCoord), g_GTAOConsts, g_srcWorkingDepth, g_samplerPointClamp, g_outWorkingAOTerm, g_outWorkingEdges );
}
// Engine-specific entry point for the second pass
[numthreads(XE_GTAO_NUMTHREADS_X, XE_GTAO_NUMTHREADS_Y, 1)]
void CSGTAOHigh( const uint2 pixCoord : SV_DispatchThreadID )
{
// g_samplerPointClamp is a sampler with D3D12_FILTER_MIN_MAG_MIP_POINT filter and D3D12_TEXTURE_ADDRESS_MODE_CLAMP addressing mode
XeGTAO_MainPass( pixCoord, 3, 3, SpatioTemporalNoise(pixCoord, g_GTAOConsts.NoiseIndex), LoadNormal(pixCoord), g_GTAOConsts, g_srcWorkingDepth, g_samplerPointClamp, g_outWorkingAOTerm, g_outWorkingEdges );
}
// Engine-specific entry point for the second pass
[numthreads(XE_GTAO_NUMTHREADS_X, XE_GTAO_NUMTHREADS_Y, 1)]
void CSGTAOUltra( const uint2 pixCoord : SV_DispatchThreadID )
{
// g_samplerPointClamp is a sampler with D3D12_FILTER_MIN_MAG_MIP_POINT filter and D3D12_TEXTURE_ADDRESS_MODE_CLAMP addressing mode
XeGTAO_MainPass( pixCoord, 9, 3, SpatioTemporalNoise( pixCoord, g_GTAOConsts.NoiseIndex ), LoadNormal( pixCoord ), g_GTAOConsts, g_srcWorkingDepth, g_samplerPointClamp, g_outWorkingAOTerm, g_outWorkingEdges );
}
// Engine-specific entry point for the third pass
[numthreads(XE_GTAO_NUMTHREADS_X, XE_GTAO_NUMTHREADS_Y, 1)]
void CSDenoisePass( const uint2 dispatchThreadID : SV_DispatchThreadID )
{
const uint2 pixCoordBase = dispatchThreadID * uint2( 2, 1 ); // we're computing 2 horizontal pixels at a time (performance optimization)
// g_samplerPointClamp is a sampler with D3D12_FILTER_MIN_MAG_MIP_POINT filter and D3D12_TEXTURE_ADDRESS_MODE_CLAMP addressing mode
XeGTAO_Denoise( pixCoordBase, g_GTAOConsts, g_srcWorkingAOTerm, g_srcWorkingEdges, g_samplerPointClamp, g_outFinalAOTerm, false );
}
[numthreads(XE_GTAO_NUMTHREADS_X, XE_GTAO_NUMTHREADS_Y, 1)]
void CSDenoiseLastPass( const uint2 dispatchThreadID : SV_DispatchThreadID )
{
const uint2 pixCoordBase = dispatchThreadID * uint2( 2, 1 ); // we're computing 2 horizontal pixels at a time (performance optimization)
// g_samplerPointClamp is a sampler with D3D12_FILTER_MIN_MAG_MIP_POINT filter and D3D12_TEXTURE_ADDRESS_MODE_CLAMP addressing mode
XeGTAO_Denoise( pixCoordBase, g_GTAOConsts, g_srcWorkingAOTerm, g_srcWorkingEdges, g_samplerPointClamp, g_outFinalAOTerm, true );
}
// Optional screen space viewspace normals from depth generation
[numthreads(XE_GTAO_NUMTHREADS_X, XE_GTAO_NUMTHREADS_Y, 1)]
void CSGenerateNormals( const uint2 pixCoord : SV_DispatchThreadID )
{
float3 viewspaceNormal = XeGTAO_ComputeViewspaceNormal( pixCoord, g_GTAOConsts, g_srcRawDepth, g_samplerPointClamp );
// pack from [-1, 1] to [0, 1] and then to R11G11B10_UNORM
g_outNormalmap[ pixCoord ] = XeGTAO_FLOAT3_to_R11G11B10_UNORM( saturate( viewspaceNormal * 0.5 + 0.5 ) );
}
///
///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////